I Shut Down My AI Agency. Here's What's Worth Building

Author: Arseny Shatokhin · Watch on YouTube

Video authored by the channel shown, not by IANIA; all rights belong to the creator. Transcript and summary generated automatically; the original-language content is preserved.

Original language: English

Original video: I Shut Down My AI Agency. Here’s What’s Worth Building · Channel: Arseny Shatokhin · 11:21 · Watch on YouTube

Topics: AI agents, Enterprise AI applications, Task and workflow automation, Claude Code, Markets and economy

Summary

Arsenii explains why he shut down his AI agency: custom client work had low margins and did not consistently deliver lasting results. He recommends replacing one-off automations with productized systems that clients can adapt themselves, and building focused software for underserved niches after validating demand and collecting deposits. He warns against building flashy AI dashboards or recreating existing platforms without a clear need. His main advice is to prioritize real relationships and accountability for outcomes, using AI behind the scenes to deliver results.

Key points

Custom automations for every client are difficult to scale; productize services once you have experience and case studies.
AI systems need to evolve as businesses change, so clients should be able to customize and maintain them themselves.
A productized AI OS install combines a reusable system with training videos and skills that help business owners adapt it.
Avoid starting with an enormous, billion-dollar vision; focus first on a small problem with a clear end user.
Niche software can deliver strong value, even with smaller models, when it is tailored to a specific underserved community.
Validate a niche product by asking potential users about their problems and willingness to pay, then consider charging a deposit before building.
Custom AI OS dashboards are not valuable just because they look impressive; the useful work is organizing context and making it actionable.
Rebuilding an established platform to save money may not justify the added maintenance burden unless a specific niche has unmet needs.
Real connections and accountability for results are advantages people can offer that AI does not provide.
Build relationships and secure commitments before building; use AI in the background to help deliver the promised outcome.

Chapters

Start	Chapter	Summary
00:00	Why the AI agency closed	Arsenii describes the agency’s low margins and inconsistent client results, then introduces what he believes is and is not worth building.
01:06	Productize client solutions	He argues against custom automations for every client and explains how a self-customizing AI OS install lets business owners adapt and maintain their systems.
04:22	Build for a focused niche	He cautions against oversized visions and recommends validating niche software with its intended users before building, potentially collecting deposits first.
07:32	Avoid building AI dashboards without a purpose	He questions the value of flashy AI OS dashboards and unnecessary platform rebuilds, emphasizing that a build should solve a clear problem.
09:09	Prioritize relationships and results	He says human connections and accountability remain valuable, and urges viewers to connect with potential clients and secure commitments before building.

Quotes

The margins were low, and the client results weren’t quite there. — 01:06

please find a service that you can productize and deliver consistently. — 01:37

niche software products for niche underserved communities. — 05:25

real connections with real people. — 09:09

stop building products and start building real connections. — 10:49

Concepts and entities

AI OS install, custom automations, niche software products, AI OS, AppointlyFi, Agency, Well-Founded Creatives, Claude Code, Beehiiv, Sam Altman, Mark Lou, Deloitte, McKinsey, Claude Opus 5.5, Arsenii

Full transcript (original language)

00:00 I shut down my AI agency about nine months ago, and in this video, I’m gonna show you what is actually worth building today. The reason I shut it down was simple. The margins were low, and the client results weren’t quite there. Since then, of course, AI has improved tremendously, and so today you can build so much more without hiring more people. But this raises another important question, what is actually worth building today? Or even more important, what is not worth building? Because I see so many people simply focusing on their own

00:31 thing, and I get asked this question constantly. So by the end of this video, you’ll know the exact three things that are not worth building today, the three things that are worth building today, and how you can get started over the weekend. And if you get this right, I guarantee that you will be able to make a steady income stream in the AI era without being constantly afraid of losing your job or not having any work in the future. But first, quickly about me. My name is Arsenii, and I’ve been building in AI and working for myself for over five years now. I started as a freelancer on Upwork,

01:06 and then I scaled my AI agency to $80,000 per month before I shut it down. Then I also scaled AppointlyFi, which is my fully automated AI lead generation company, to $15,000 per month, and my open source projects also have 7,000 stars on GitHub. Now let’s start with the number one thing that’s not worth building anymore, which is custom automations for every client if you’re still building custom automations agents or other solutions for every single client, , I urge you to stop doing this as soon as you can.

01:37 This is completely fine at the beginning for your first one or two projects, but then as soon as you get some case studies and as soon as you get some experience in a certain niche, please find a service that you can productize and deliver consistently. Today, AI is such an integral part of every single business that you cannot simply outsource AI development to some team overseas. It would be kinda like hiring contractors for some of the most key roles at your company. Contractors simply don’t know how your company runs, and if you don’t have a clear process

02:10 for hiring and training employees, well, your business is just never gonna scale. and so that’s what we noticed in our own AI agency as well, is that we would build an AI agent for a business, and then they would start using it. They would be excited about it for a week or two, and then they would simply stop, and they would either switch to something else or they would stop using it completely. And the reason is because an agent is never perfect. There is no perfect skill, there is no perfect agent, and there is no perfect automation that you can build for a business. every single business evolves over time.

02:43 And so every agent that you build for that business must evolve with it as well So here’s what actually got my clients results. Now I sell a productized offer that customizes itself. Essentially, now I offer this AI OS install, which is the exact system that I built myself first in my own business over the last four years, and then I provide training videos and AI skills that allow business owners to customize the same system themselves. So this solves two issues. Number one, it solves my time, and I don’t have to

03:17 hire anyone to deliver it because the business owners literally customize it themselves. They watch the video, then they run the skill. The skill asks them certain questions, and then based on these questions, this skill either builds new agents, new skills, or customizes the system for them specifically. And secondly, they now understand how to actually run the system themselves. so if there’s some adjustment or something that isn’t perfect yet, they don’t have to come back to me every single time, and I don’t have to do the work for them.

03:48 They can simply adjust it themselves, and this way they keep evolving their own agents and AI systems as their company grows. And this has completely changed everything for me. The client results that I’m getting right now from these productized AI OS installs are just on a completely different level from what we used to get from custom solutions and I don’t even have to do any kind of outreach simply because clients refer me new clients,I have two case studies on my website already, and I have two more coming soon. Okay, now if building is so cheap, another trap that I see people fall into is going too big,

04:22 and I actually personally fell for that myself And even Sam Altman said that you can now build all thirty things, and you can get people to use them all. So today, you really don’t have to be laser-focused on one thing to make it successful. You can build three, four things at the same time, although I certainly do not recommend this at the beginning. But where most people fall short is instead of focusing on a small solution that people really need, they focus on this billion-dollar problem even before they made their first 10K online. And

04:55 I actually felt for the same trap myself as well. I was trying to build a platform called Agency, which is a billion-dollar vision, and unfortunately, for now it didn’t work. for now, I paused this project. I don’t think it will be over yet, but I’m currently trying to find product market fit in a different niche, which is exactly what I’m gonna talk about next. And so what I’m doing now is instead I’m focusing on more niche use cases where it’s really clear who the end user is

05:25 So the thing worth building number two is niche software products for niche underserved communities. While everyone is chasing this billion-dollar vision and everyone is going after the same thing, which is kinda obvious today, you can serve a very specific niche market that no one else is chasing right now. And that’s exactly what I’m doing right now with my own framework and the platform. Instead of building a product for everyone, which basically means for no one, I’m focusing on one specific

05:56 community called Well-Founded Creatives, and we’re building a very specific set of agents for them. And what I’m seeing is that the feedback, kinda like with custom solutions versus productized offerings, is just so much better and so much clearer. Even with smaller models, even with not fully fine-tuned agents, they are already getting a ton of value because these agents are made specifically for them. They help them to create their decks, nail down their messaging, and get to the right offer a lot faster.

06:27 what also surprised me is that because these agents are so tightly fine-tuned for this specific community on these specific resources for these people, we also don’t even have to use these latest crazy frontier models. They get results even with smaller ones just as well But the best part is that because you’re building for the specific niche market or a niche community, you don’t have to build anything upfront. You can simply ask them what is their problem and how much would they pay to get this problem solved. And then you can

06:59 charge a deposit and promise them that you’re gonna help them to get this problem solved, and only then you actually build the product and then charge the full price. This way, you de-risk yourself from building something for three months that eventually no one even needs However, even when you do know who this is for, there is one more trap that a lot of people fall into. Building things that consume your usage but do not move any of your goals forward. a clear example of such a thing and what is not worth building, number three, is your own custom AI OS.

07:32 So I have seen a ton of people vibe code their own AI OS dashboards that look like this with all these lines and moving parts. They look amazing, and I would love to play with one myself. But the question is, what problem does it solve? Like, what, what does it actually do? What is on this dashboard right now? that I cannot get from Claude Code? What actually matters when you’re building your AI OS is the structure. It’s organizing all your context and making it actionable for your AI, not connecting some random dots and lines together. That literally does not bring any

08:06 value Here is another example that I recently saw on Twitter. Mark Lou, a very famous indie hacker, recently vibe coded his own Beehive to go from about $4,000 a year to under $300 a year. And Beehive obviously worked for him, but he wanted to cut costs, and that’s what he did. And the founder of Beehive also posted all of the features that you would need, which is obviously like a crazy amount of things that aren’t even related to vibe coding and that you can’t get just from maxing out your usage on CloudCode.

08:39 Like for example, 99% deliverability and avoiding spam filters If he built it for some niche community where some features in Beehiiv were clearly missing, then I would understand that. But rebuilding an entire platform because you want to cut 3.5K per year and adding all that maintenance depth on top just doesn’t make sense to me So what is actually worth building? The last thing and the most important thing, number three, that is worth building is real

09:09 connections with real people. Two things that AI will never be able to automate are, number one, building connections with people, because people build connections with other people, not with AI. And number two, taking accountability for the results, because whether you get the result or not, you’re gonna have to pay for the tokens. AI does not guarantee you any kinds of results. It simply charges you per token. And so of course, there are no molds today. You can rebuild pretty much anything. But if you have built a specific connection with someone who is already successful in their

09:43 own niche, and if they know that you care about the results that they’re gonna get, they will still choose to work with you instead of working with AI This is why consulting firms like Deloitte and McKinsey still exist. They don’t necessarily do anything better than Claude Opus 5.5, for example, but many Fortune 500 companies still go to them whenever something happens for two reasons. Number one, they simply already have the connections with those consulting firms, and it’s kinda like automatic for them.

10:14 It’s almost in their SOPs, And number two is because they actually take responsibility for the outcomes of the case. they don’t just hand you over a document, but if the deal falls through because of this document, because it’s incorrect or something, they’re actually gonna reimburse you for all the expenses and the costs. So again, this is not something that AI is gonna be able to automate anytime soon. This is where people will still be needed whatever happens next. Even if models become smarter than all humans combined, you will still need someone to blame for the outcomes of the actions. And so if you wanna stand out today in 2026, the most important thing is to build real

10:49 connections with real people and guaranteeing a real result. And then you use AI on the back end yourself to deliver that result. And I know it feels very tempting to keep building. I fall for this all the time myself as well. But unfortunately, most of the time it’s the least productive thing you can do. So what you should do instead is stop building products and start building real connections. Reach out to a few people first. And then preferably even charge upfront before you build anything. And if you wanna see exactly how to do this, how to actually connect with someone and send an outreach message live, I recorded my entire LinkedIn outbound system in this video. I show you everything, how to get to the right offer, how to write your messaging, how to send an actual message that cuts through all the AI slop.

11:21 I actually even record a message live and then send it during this video. So check it out, it’s completely free, and I’ll see you there